back to solo canny

app.py

@@ -15,7 +15,10 @@ from diffusers.utils import load_image
 
 # load pipeline
 controlnet_canny = SD3ControlNetModel.from_pretrained("InstantX/SD3-Controlnet-Canny")
-controlnet_tile = SD3ControlNetModel.from_pretrained("InstantX/SD3-Controlnet-Tile")
+pipe = StableDiffusion3ControlNetPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-3-medium-diffusers",
+    controlnet=controlnet_canny
+).to("cuda", torch.float16)
 
 def resize_image(input_path, output_path, target_height):
     # Open the input image
@@ -36,47 +39,21 @@ def resize_image(input_path, output_path, target_height):
 
     return output_path, new_width, target_height
 
-def show_hidden():
-    return gr.update(visible=True)
-
-def load_pipeline(control_type, progress=gr.Progress(track_tqdm=True)):
-    global pipe_canny, pipe_tile
-    if control_type == "canny":
-        global pipe_canny
-        pipe_canny = StableDiffusion3ControlNetPipeline.from_pretrained(
-            "stabilityai/stable-diffusion-3-medium-diffusers",
-            controlnet=controlnet_canny
-        )
-    elif control_type == "tile":
-        global pipe_tile
-        pipe_tile = StableDiffusion3ControlNetPipeline.from_pretrained(
-            "stabilityai/stable-diffusion-3-medium-diffusers",
-            controlnet=controlnet_tile
-        )
-    return gr.update(value="pipeline ready", visible=True)
 
 @spaces.GPU(duration=90)
-def infer(image_in, prompt, control_type, inference_steps, guidance_scale, control_weight, progress=gr.Progress(track_tqdm=True)):
-    global pipe, pipe_canny, pipe_tile
+def infer(image_in, prompt, inference_steps, guidance_scale, control_weight, progress=gr.Progress(track_tqdm=True)):
+
     n_prompt = 'NSFW, nude, naked, porn, ugly'
 
-    if control_type == "canny":
-        pipe = pipe_canny
-        pipe.to("cuda", torch.float16)
-        # Canny preprocessing
-        image_to_canny = load_image(image_in)
-        image_to_canny = np.array(image_to_canny)
-        image_to_canny = cv2.Canny(image_to_canny, 100, 200)
-        image_to_canny = image_to_canny[:, :, None]
-        image_to_canny = np.concatenate([image_to_canny, image_to_canny, image_to_canny], axis=2)
-        image_to_canny = Image.fromarray(image_to_canny)
-
-        control_image = image_to_canny
-    
-    elif control_type == "tile":
-        pipe = pipe_tile
-        pipe.to("cuda", torch.float16)
-        control_image = load_image(image_in)
+    # Canny preprocessing
+    image_to_canny = load_image(image_in)
+    image_to_canny = np.array(image_to_canny)
+    image_to_canny = cv2.Canny(image_to_canny, 100, 200)
+    image_to_canny = image_to_canny[:, :, None]
+    image_to_canny = np.concatenate([image_to_canny, image_to_canny, image_to_canny], axis=2)
+    image_to_canny = Image.fromarray(image_to_canny)
+
+    control_image = image_to_canny
  
     # infer
     image = pipe(
@@ -88,15 +65,11 @@ def infer(image_in, prompt, control_type, inference_steps, guidance_scale, contr
         guidance_scale=guidance_scale,
     ).images[0]
 
-    if control_type == "canny":
-
-        image_redim, w, h = resize_image(image_in, "resized_input.jpg", 1024)
-        image = image.resize((w, h), Image.LANCZOS)
     
-        return image, gr.update(value=image_to_canny, visible=True)
+    image_redim, w, h = resize_image(image_in, "resized_input.jpg", 1024)
+    image = image.resize((w, h), Image.LANCZOS)
     
-    elif control_type == "tile":
-        return image, gr.update(value=None, visible=False)
+    return image, gr.update(value=image_to_canny, visible=True)
    
 
 css="""
@@ -111,6 +84,7 @@ with gr.Blocks(css=css) as demo:
         # SD3 ControlNet
 
         Experiment with Stable Diffusion 3 ControlNet models proposed and maintained by the InstantX team.<br />
+        Model card: (InstantX/SD3-Controlnet-Canny)[https://huggingface.co/InstantX/SD3-Controlnet-Canny]
         """)
         
         with gr.Column():
@@ -119,14 +93,7 @@ with gr.Blocks(css=css) as demo:
                 with gr.Column():
                     image_in = gr.Image(label="Image reference", sources=["upload"], type="filepath")
                     prompt = gr.Textbox(label="Prompt")
-                    control_type = gr.Radio(
-                        label="Control type",
-                        choices = [
-                            "canny",
-                            "tile"
-                        ],
-                        value="canny"
-                    )
+                    
                     with gr.Accordion("Advanced settings", open=False):
                         with gr.Column():
                             with gr.Row():
@@ -137,23 +104,14 @@ with gr.Blocks(css=css) as demo:
                     submit_canny_btn = gr.Button("Submit")
                     
                 with gr.Column():
-                    models = gr.Textbox(label="Plug-in pipes", visible=False)
                     result = gr.Image(label="Result")
                     canny_used = gr.Image(label="Preprocessed Canny", visible=False)
                     
 
 
     submit_canny_btn.click(
-        fn = show_hidden,
-        inputs = None,
-        outputs = [models]
-    ).then(
-        fn = load_pipeline,
-        inputs = [control_type],
-        outputs = [models]
-    ).then(
         fn = infer,
-        inputs = [image_in, prompt, control_type, inference_steps, guidance_scale, control_weight],
+        inputs = [image_in, prompt, inference_steps, guidance_scale, control_weight],
         outputs = [result, canny_used],
         show_api=False
     )